Data-visualization critique and design with Tufte principles

A Claude Code skill that applies Edward Tufte's data-visualization principles both to design new charts and to critique existing ones. It supplies evaluation checklists rather than code: a 7-question Tufte Test, a 14-question Extended Tufte Test, the Lie Factor calculation, and the Eraser Test, plus reference docs on data-ink ratio, chartjunk elimination, graphical integrity, small multiples, sparklines, and information density.

Most chart review stops at "looks fine." This replaces that with a checklist and a falsifiable integrity metric — the Lie Factor (size of effect shown in the graphic ÷ size of effect in the data; 1.0 = truthful). It catches misleading axes, decorative chartjunk, and low information density before a chart ships in a board deck, investor update, or marketing report.

1. Install the skill. Fetch its SKILL.md plus the two reference docs into .claude/skills/.
2. For a new visualization: clarify the data story and audience → pick the viz type → design with the minimum necessary elements → run the evaluation checklist before shipping.
3. For a critique: compute the Lie Factor to verify graphical integrity → list decorative/chartjunk elements → assess the data-ink ratio → output specific, ranked improvements.
4. Gate on the Tufte Test (7 questions) or the Extended Tufte Test (14 questions) as a publish check.

• Checklist, not code — it guides judgment, it doesn't render charts; pair with your existing charting tool.
• Critique quality depends on the agent actually seeing the chart — paste the image, not just a description.

• Claude Code
• The skill files (SKILL.md + two reference docs), installable from the public gist:

# fetch SKILL.md + references from the gist into.claude/skills/tufte-viz/
https://gist.github.com/aparente/e48c353755958621b3c0004593105a90

• Image input when critiquing an existing chart (the agent must see the chart)